TWI287702B - Managing system power - Google Patents
Managing system power Download PDFInfo
- Publication number
- TWI287702B TWI287702B TW091135992A TW91135992A TWI287702B TW I287702 B TWI287702 B TW I287702B TW 091135992 A TW091135992 A TW 091135992A TW 91135992 A TW91135992 A TW 91135992A TW I287702 B TWI287702 B TW I287702B
- Authority
- TW
- Taiwan
- Prior art keywords
- variable
- value
- spare capacity
- resource
- parameter
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
- G06F1/26—Power supply means, e.g. regulation thereof
- G06F1/32—Means for saving power
- G06F1/3203—Power management, i.e. event-based initiation of a power-saving mode
- G06F1/3234—Power saving characterised by the action undertaken
- G06F1/324—Power saving characterised by the action undertaken by lowering clock frequency
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
- G06F1/26—Power supply means, e.g. regulation thereof
- G06F1/32—Means for saving power
- G06F1/3203—Power management, i.e. event-based initiation of a power-saving mode
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Power Sources (AREA)
Abstract
Description
1287702 玫、發明說明 尤月應敘明：發明所屬之技術領域、先前技術、内容、實施方式及圖式簡單說明） 【發明所屬之技術領域】 發明領域 本發明係有關系統電源，更具體地係有關系統電源管 5 理。 【先前技術；1 發明背景 電腦系統包括使用電極板的網路伺服器通常只要一開 啟電源就會全程以全部電源運作，並不會考量到傳送運算 1 〇服務至終使用者所需計算量之電腦負載。電腦負載包括 如中央處理單元（CPU)週期、磁碟存取、記憶體存取、輸 入-輸出（I/O)存取等等。膝上型電腦以及以電池作為電力 的系統具有各種不同的運作模式，例如“睡眠”、“待機，，' 冬眠等等，當系統沒有在運作時用以降低電源。然而這 些能源節省模式通常是依據系統是否在使用中，亦即運= 或非運作中來決定，並不是依據系統效能或系統負栽而定 。再者’假若有任何使用者動作，即使此動作並不需要全 部的電源，這些模式仍舊會將系統切換至全速運作:因^ 伺服器很少完全處於非動作狀態，他們大部分時間都是全 20速運作但並不會比膝上型電腦或以電池為電力的系統^ 源節省模式得到更大的助益。根據上述論點，有需要 -機構用以解決上述的缺陷以及相關的問題。 供 C 明内】 發明概要 1287702 玖、發明說明 本發明在各個實施例令提供管理系統電源用之技術。 在-貫施例中，在系統上運作的服務導致系統計算負載以 及/或系統資源消耗電源。為了更有效的管理能源的消耗 ’會週期性的估量備用的系統資源容量，當此備用容量超 5過-預定的範圍時，將會調整資源運算，例如增加或減慢 CPU速度，使得備用容量在此預定範圍内。再者，備用容 5貫際實際上盡可能維持接近零，而且此備用容量是根據 週期性估量而得的資源使用值統計分佈來決定。此備用容 量也會根據系統資源飽和的機率加以計算。 10 在一實施例中，為了維持服務層級協定（SLA)所需的 服務，會決定SLA參數與資源使用之間的關係。除了其他 因素以及參數之間的關連外，資源使用的備用容量會根據 SLA參數的備用容量加以調整。 各種實施例包括在計算系統備用容量前將系統效能最 15佳化、節省系統群組或叢集的電源、在特殊狀況如電壓不 足、高溫時節省電源等等。 圖式簡單說明 本發明是以範例的方式作說明，但並不侷限於所附圖 示中參考標號對照之類似元件，這些圖示為： 20 第1圖展示根據本發明實施例可能實現的網路。 第2圖展示根據本發明實施例可能實現的示範電腦。 第3A圖展示在給定參數m時用以找出機率之對照表。 第3B圖展示在給定機率尸時用以找出參數所之對照表。 第4圖展示兩個變數X與少之走勢曲線 1287702 玖、發明說明 第5 A圖為解說依據一實施例管理系統電源步驟之流程 圖。 第5B圖為解說依據一實施例管理與SLA參數相關之系 統電源步驟之流程圖。 第6圖展示依據一實施例之電源管理。 第7圖為解說依據一實施例執行電源管理之流程圖。 t實施方式】 較佳實施例之詳細說明 網路概觀 10 在下面的說明中，為了方便說明，會提出許多特定部 分的詳細說明以便對本發明能有透徹的瞭解。然而本發明 可以以這些特定說明以外，在業界熟知的技藝加以實現。 另一方面，熟知架構以及裝置是以方塊圖形式展示，以避 免與本發明部分混淆。 15 第1圖為根據本發明實施例可能實現之網路1 〇〇。網路 100包括伺服器110、數個客戶端120-1、120-2、…，120-N 、以及通訊鏈結150。在一實施例中，商業機構使用網路 1〇〇提供他們的客戶電腦服務，其中伺服器11〇經由客戶端 系統120提供客戶服務。一般而言，這些機構以及他們的 2〇客戶以服務層級協定（SLA)為基礎來對所提供的服務設定 條件，例如在某特定時間内伺服器11〇服務已提出的請求 、驗證數個使用者的身份、提供特定數量的網頁等等。網 路100在本文件中是作為範例，各種變化亦包含在本發明 範缚内。例如Η司服器110可以自行獨立而不需要位於通 1287702 玖、發明說明 訊連結150上；數個伺服器11〇可以以叢集方式運作或是以 負載平衡方式運作；網路100可以為各種網路協定等等。 伺服器110可以是任何電腦並執行各種服務，例如網 頁、資料庫、郵件、安全、通訊、會計、負載平衡、檔案 5儲存服務等等。此服務清單是做說明用；由伺服器110執 行的其他服務如程式、負載以及類似的服務亦在本發明範 臂内。為了請求服務，使用者或應用程式從客戶端1 戋 伺服器（圖中並未展示）經由通訊連結15〇傳送請求至伺服器 110 〇 10 通訊連結150為伺服器110與客戶端通訊的機構。通訊 連結150可以是單一網路或使用一種或多種通訊協定如傳 輸控制協定/網際網路協定（TCP/IP)、大眾交換式電話網路 (PSTN)、數位用戶線路（DSL)、纜線網路、衛星、無線網 路等等組合的複合式網路。通訊連結15〇的範例包括網路 15媒體互相連接結構、環形連接、橫向連接等等。每一客 戶端120可以使用不同的通訊連結與伺服器110連結。在一 實施例中，通訊連結150為網際網路。 電腦系統概觀 第2圖為展示一句本發明實施例可能實現之電腦系統 20 200之方塊圖。舉例來說，電腦系統2〇〇可以實作為伺服器 n〇、客戶端120等等。在一實施例中，電腦系統2〇〇包括 中央處理單元（CPU)204、隨機存取記憶體（RAM)2〇8、唯 讀記憶體（R〇M)212、儲存裝置216以及通訊界面22〇，上 述所有元件皆連接至匯流排224。 1287702 玖、發明說明 CPU 204控制邏輯、程序資訊、以及協調電腦系統200 内的活動。一般而言，CPU 204執行儲存於RAM 208與 ROM 212中的指令，例如協調資料從輸入裝置228移動至 顯示裝置232。CPU 204可能包括一或多個處理器。 5 RAM 208通常稱之為主記憶體或記憶體系統，用以暫 時地儲存CPU 204要執行的資訊與指令。RAM 208的形式 可能為單排記憶體模組（SIMM)或雙排記憶體模組（DIMM) 。在RAM 208中的資訊可能從輸入裝置228取得或是由 CPU 204產生來給CPU204執行指令時所需要的部分演算程 10 序。 ROM 212儲存資訊與指令，資料一旦寫入ROM晶片中 便只能讀取而無法更改或移除。在一實施例中，ROM 212 儲存電腦系統200組配與初始運作之指令。 通訊界面220使得電腦系統200可以與其他電腦或裝置 15 溝通。通訊界面220可能為數據機、整合服務數位網路 (ISDN)界面卡、區域網路（LAN)通訊埠等等。這些業界技 藝會認定數據機或ISDN界面卡式經由電話線提供資料通 訊，而LAN通訊埠是經由LAN提供資料通訊。通訊界面 220亦允許無線通訊。 20 匯流排224可以是任何通訊機構，用以通訊電腦系統 200使用的資訊。在第2圖的範例中，匯流排224為在CPU 204、RAM 208、ROM 212、儲存裝置 216、通訊界面 220 等等之間傳輸資料之媒介。 電腦系統200典型地連接至輸入裝置228、顯示裝置 1287702 玖、發明說明 232、以及游標控制236。輸入裝置228，例如鍵盤包括字 母與數字以及其他按鍵，傳遞資訊與指令至CPU 204。顯 示裝置232，例如陰極射線管（CRT)，顯示資訊至電腦系統 2 0 0的使用者。游標控制2 3 6，例如滑鼠、執跡球、或游標 5 方向鍵，傳地方向資訊與指令至CPU 204並且控制顯示裝 置232上游標的移動。 電腦系統200可以透過一或多個網路與其他電腦或裝 置通訊。舉例來說，電腦系統200使用通訊界面220經由網 路240與另一台與印表機248連接之電腦通訊，或者經由全 10 球資訊網252與伺服器256通訊。全球資訊網252通常稱之 為“網際網路’’。換言之，電腦系統200可以經由網路240存 取網際網路252。 電腦系統200可以用來實現此處揭露之技術。在各個 實施例中，CPU 204藉由執行傳送到RAM208之指令來執 15 行此技術的各個步驟。在替代的實施例中，實體線路電路 可以用來取代或與軟體指令結合來實現上述技術。因此， 本發明的實施例並不侷限於任一個韌體、軟體、硬體或電 路或是其組合。1287702 玫,发明说明,尤月应述: The technical field, prior art, content, embodiment and schematic description of the invention belong to the technical field of the invention. FIELD OF THE INVENTION The present invention relates to system power supplies, and more particularly About the system power supply. [Prior Art; 1 BACKGROUND OF THE INVENTION Computer systems, including network servers using electrode plates, usually operate at full power as long as the power is turned on, and do not take into account the amount of computation required to transfer the operation 1 to the end user. Computer load. Computer loads include, for example, central processing unit (CPU) cycles, disk access, memory access, input-output (I/O) access, and more. Laptops and battery-powered systems have different modes of operation, such as "sleep," "standby," hibernate, etc., to reduce power when the system is not operating. However, these energy-saving modes are usually Depending on whether the system is in use, whether it is in operation or not, it is not based on system performance or system load. In addition, if there is any user action, even if this action does not require all power, These modes will still switch the system to full speed: because the servers are rarely completely inactive, they spend most of their time running at 20 speeds but not on laptops or batteries. The source saving mode is more beneficial. According to the above arguments, there is a need for an organization to solve the above-mentioned defects and related problems. For the purpose of the invention, the invention is summarized as follows: 1. The invention is provided in various embodiments. Technology used in system power. In the example, services running on the system cause the system to calculate the load and/or system resource consumption. Power supply. In order to manage energy consumption more effectively, the spare system resource capacity will be periodically estimated. When the spare capacity exceeds 5 - the predetermined range, resource operations will be adjusted, such as increasing or slowing down the CPU speed. The spare capacity is within this predetermined range. Furthermore, the standby capacity is actually maintained as close to zero as possible, and the spare capacity is determined based on the statistical distribution of the resource usage values obtained by the periodic estimation. The probability of system resource saturation is calculated. 10 In an embodiment, in order to maintain the services required by a service level agreement (SLA), the relationship between SLA parameters and resource usage is determined, among other factors and the relationships between the parameters. The spare capacity for resource usage is adjusted based on the spare capacity of the SLA parameters. Various embodiments include maximizing system performance before computing system backup capacity, saving system groups or clusters of power, in special conditions such as insufficient voltage, Saving power at high temperatures, etc. The drawings briefly illustrate the present invention by way of example, but not Similar elements in the accompanying drawings are referenced to the like elements, and these illustrations are: 20 Figure 1 shows a network that may be implemented in accordance with an embodiment of the present invention. Figure 2 shows an exemplary computer that may be implemented in accordance with an embodiment of the present invention. Figure 3A shows a comparison table used to find the probability given a parameter m. Figure 3B shows a comparison table used to find the parameters at a given probability. Figure 4 shows the trend of two variables X and less. Curve 1287702 发明, INSTRUCTION DESCRIPTION FIG. 5A is a flow chart illustrating the steps of managing a system power supply in accordance with an embodiment. FIG. 5B is a flow chart illustrating the steps of managing system power associated with SLA parameters in accordance with an embodiment. Power management according to an embodiment. Fig. 7 is a flow chart illustrating the execution of power management according to an embodiment. t Embodiments Detailed Description of the Preferred Embodiments Network Overview 10 In the following description, for convenience of explanation, A detailed description of many specific parts is set forth to provide a thorough understanding of the invention. However, the invention may be practiced otherwise than as specifically described in the art. On the other hand, well-known architectures and devices are shown in block diagram form in order to avoid confusion with the present invention. 15 Figure 1 shows a network 1 that may be implemented in accordance with an embodiment of the present invention. The network 100 includes a server 110, a plurality of clients 120-1, 120-2, ..., 120-N, and a communication link 150. In one embodiment, the business organization uses the network to provide their client computer services, wherein the server 11 provides customer service via the client system 120. In general, these organizations and their 2 customers are based on service level agreements (SLAs) to set conditions for the services provided, such as requests made by the server 11 service at a certain time, and verification of several uses. Identity, provide a specific number of web pages, and more. The network 100 is exemplified in this document, and various changes are also included in the scope of the present invention. For example, the server 110 can be independent of itself and does not need to be located on the 1287772 玖, invention description link 150; several servers 11 〇 can operate in a cluster mode or in a load balancing manner; the network 100 can be a variety of networks Road agreement and so on. Server 110 can be any computer and performs various services such as web pages, databases, mail, security, communications, accounting, load balancing, file storage services, and the like. This list of services is for illustrative purposes; other services such as programs, loads, and the like that are performed by server 110 are also within the scope of the present invention. In order to request the service, the user or application transmits a request from the client 1 伺服 server (not shown) via the communication link 15 to the server 110 〇 10 The communication link 150 is a mechanism for the server 110 to communicate with the client. The communication link 150 can be a single network or use one or more communication protocols such as Transmission Control Protocol/Internet Protocol (TCP/IP), Mass Switched Telephone Network (PSTN), Digital Subscriber Line (DSL), cable network. A hybrid network of roads, satellites, wireless networks, and more. Examples of communication links include network 15 media interconnect structures, ring connections, horizontal connections, and the like. Each client 120 can be coupled to the server 110 using a different communication link. In one embodiment, the communication link 150 is an internet connection. Overview of Computer System FIG. 2 is a block diagram showing a computer system 20 200 that may be implemented in an embodiment of the present invention. For example, the computer system 2 can be implemented as a server n, a client 120, and the like. In one embodiment, the computer system 2 includes a central processing unit (CPU) 204, a random access memory (RAM) 2〇8, a read-only memory (R〇M) 212, a storage device 216, and a communication interface 22 That is, all of the above components are connected to the bus bar 224. 1287702 发明, Invention Description The CPU 204 controls logic, program information, and coordinates activities within the computer system 200. In general, CPU 204 executes instructions stored in RAM 208 and ROM 212, such as coordination material moving from input device 228 to display device 232. CPU 204 may include one or more processors. 5 RAM 208 is commonly referred to as a main memory or memory system for temporarily storing information and instructions to be executed by CPU 204. The RAM 208 may be in the form of a single-row memory module (SIMM) or a dual-row memory module (DIMM). The information in RAM 208 may be retrieved from input device 228 or generated by CPU 204 to provide a portion of the programming sequence required by CPU 204 to execute the instructions. The ROM 212 stores information and instructions which, once written into the ROM chip, can only be read and cannot be changed or removed. In one embodiment, ROM 212 stores instructions for computer system 200 to be assembled and initially operational. Communication interface 220 allows computer system 200 to communicate with other computers or devices 15. The communication interface 220 may be a data machine, an integrated services digital network (ISDN) interface card, a local area network (LAN) communication port, and the like. These industry experts recognize that a data modem or ISDN interface card provides data communication via a telephone line, while a LAN communication device provides data communication via a LAN. Communication interface 220 also allows for wireless communication. 20 Bus 224 can be any communication mechanism used to communicate information used by computer system 200. In the example of FIG. 2, bus 224 is the medium for transferring data between CPU 204, RAM 208, ROM 212, storage device 216, communication interface 220, and the like. Computer system 200 is typically coupled to input device 228, display device 1287702, invention description 232, and cursor control 236. Input device 228, such as a keyboard, includes letters and numbers, as well as other keys, to communicate information and instructions to CPU 204. A display device 232, such as a cathode ray tube (CRT), displays information to a user of the computer system 2000. The cursor control 2 3 6, such as a mouse, a trackball, or a cursor 5 direction key, transmits direction information and commands to the CPU 204 and controls the movement of the upstream of the display device 232. Computer system 200 can communicate with other computers or devices over one or more networks. For example, computer system 200 communicates with another computer connected to printer 248 via network 240 using communication interface 220, or with server 256 via a full 10-ball information network 252. World Wide Web 252 is commonly referred to as "Internet." In other words, computer system 200 can access Internet 252 via network 240. Computer system 200 can be used to implement the techniques disclosed herein. In various embodiments The CPU 204 performs the steps of this technique by executing instructions transferred to the RAM 208. In an alternate embodiment, the physical line circuit can be used in place of or in combination with the software instructions to implement the techniques described above. Embodiments are not limited to any one of firmware, software, hardware or circuitry or a combination thereof.
由CPU204執行的指令可以儲存在一或多個電腦可讀 20 取媒體或經由此媒體攜帶，其中該媒體是指電腦可以從上 面讀取資訊之媒體。電腦可讀取媒體可能為磁碟片、硬碟 、ZIP磁碟匣、磁帶或其他磁性媒體、CD-ROM、CDRAM 、DVD-ROM、DVD-RAM或其他光學媒體、紙帶、打洞卡 或其他具有穿孔樣本之實際媒體、RAM、ROM、EPROM 1287702 玖、發明說明 或其他記憶體晶片或記憶匣。電腦可讀取媒體也可以是同 軸電纜、銅線、光纖、聲波或電磁波等等。舉例來說， CPU 204要執行的指令為一或多個韌體或軟體程式的形式 ，並且原本是儲存於經由匯流排224與電腦系統2〇〇溝通之 5 CD-ROM。電腦系統2〇〇將這些指令載入RAM 2〇8，執行 某些指令並且經由數據機與電話線之通訊界面22〇傳送某 些指令至網路，例如網路240，或網際網路252等等。遠端 電腦經由網路電腦接收資料，執行所接收到的指令並傳送 這些資料至電腦系統200以儲存於儲存裝置。 10 在一實施例中，伺服器110是實現為一台電腦200，其 包括負責管理伺服器電源系統之電源管理者280。管理者 280是以在伺服器上執行的軟體套件來實現。然而管理者 280可以在連接至網路100的其他電腦上執行。為了方便說 明，電源管理者280在其執行時是展示於記憶體2〇8中。 15 消耗電源之資原 伺服器110中的各個元件，包括cpu 2〇4、記憶體2〇8 、儲存裝置216、連接週邊用的界面卡、週邊裝置如印表 機與外接磁碟機等等，通常消耗可觀的電源。然而每一個 元件提供可以官理電源的界面。依據該元件、界面以及製 20迨商等專，此界面可能為硬體、韌體、軟體、與韌體驅動 程式以及/或軟體一同運作的硬體等等。這些技藝在業界 會認為韌體可以由其他韌體或軟體程式取代。在一實施例 中，官理者280會根據此界面的指令與規格來控制這些元 件進而控制他們的電源消耗。 1287702 玖、發明說明 由CPU 204消耗的總電源視多個參數而定，例如處理 器執行時的頻率、功率週期、系統中運作的處理器數量等 等。功率週期為處理器以正常速度執行除以總時間所得的 時間。一般而言，在正常的工作條件下，為了傳送最大的 5 計算能力，伺服器110的CPU 204通常會以最高頻率運作。 然而依照本發明的技術，當不需要最大計算能力時，管理 者280降低以及/或調整上述一或多個參數來降低CPU 204 以及伺服器110消耗的電源。舉例來說，在一實施例中， CPU 204是以位於加州聖克拉拉之Intel公司製造的 10 Pentium 4來實現，此CPU包含溫度控制電路（TCC)界面。 為了控制CPU204的速度以及其電源消耗，管理者280遵循 CPU的規格與指令透過TCC界面來改變CPU204中任一個處 理器的功率週期。管理者280亦會在適當的時機將此功率 週期設為零，也就是有效地將CPU 204停止，儘管CPU還 15 是在運作狀態。在另一實施例中，管理者280透過使用 ACPI界面的應用程式來關閉由Intel處理器晶片與主機板設 計所實現的CPU 204。 在一實施例中，RAM 208在資料置換至磁碟機時會以 慢速的RAM更新與冬眠模式運作。為了降低電源消耗，管 20 理者280會將RAM208設定成上述兩個模式之一，因為上述 任一個模式比正常運作模式需要較少的電力。磁碟機會實 現為儲存裝置216，其亦包括管理者280用以降低磁碟機轉 速使之以低電力模式運作之界面。在此模式下，磁碟片停 止旋轉。 1287702 玖、發明說明 在一實施例中，管理者280使用進階組配電源界面 (ACPI)標準來降低伺服器110的電源消耗。在8〇狀態下， 伺服器110以全速電力正常運作。在S1狀態下，管理者28〇 將CPU 204停止並更新RAM 208使伺服器110以低電力模式 · 5運作。在S2狀態下，管理者將CPU 204設定為‘‘無電源，， 模式、更新RAM 208，使伺服器11〇以比}狀態更低的電力 模式運作。在S3模式下，管理者280將CPU 204設定為‘‘無 電源”狀態、RAM208為低速更新狀態以及伺服器11〇的電 · 源供應器為低電源模式。在S4或冬眠狀態下，管理者280 10完全地關閉硬體並將系統記憶體儲存至磁碟。在S5狀態下 ，管理者280完全地關閉硬體、關閉伺服器11〇執行的作業 系統，伺服器110需要重新啟動才能回復正常運作模式。 依據系統元件負載並與應用程式無關之電源管理 在一實施例中，為了決定系統效能，管理者280並不 15 考慮在伺服器11〇上執行的應用程式來測量系統負載之系 統元件使用率如CPU、記憶體、磁碟機、匯流排等等。再 · 者，CPU使用率以％(100*(總時間-閒置時間）/總時間）表示 ;磁碟與記憶體I/O使用率以％(100*(最高速率-實際速率）/ 最高速率）表示，以及匯流排使用率以％((1〇〇*目前匯流排 20 資料速率）最高資料速率）表示。 · 為了方便說明，CPU效能用以作為範例說明。然而說 · 明的觀念亦可以使用於伺服器110的其他元件以及下面討 論的效能參數上。在一實施例中，管理者280使用作業系 統提供的應用程式可程式化界面如Microsoft Windows作 13 1287702 玖、發明說明 業系統的Win32 API來取得使用率值。管理者彻接著使用 此取得的使料值以下列公式來計算cpu使㈣的平均值 與標準差： 1 i=nThe instructions executed by the CPU 204 can be stored in or carried by one or more computer readable media, wherein the media refers to media from which the computer can read information. Computer readable media may be floppy disks, hard drives, ZIP drives, magnetic tape or other magnetic media, CD-ROM, CDRAM, DVD-ROM, DVD-RAM or other optical media, tape, card or Other actual media, RAM, ROM, EPROM 1287702, invention instructions or other memory chips or memory cartridges with perforated samples. Computer readable media can also be coaxial cables, copper wires, fiber optics, sound waves or electromagnetic waves. For example, the instructions to be executed by the CPU 204 are in the form of one or more firmware or software programs, and are originally stored in a 5 CD-ROM that communicates with the computer system 2 via the bus 224. The computer system 2 loads these instructions into the RAM 2〇8, executes certain instructions and transmits certain commands to the network via the communication interface 22 of the data plane and the telephone line, such as the network 240, or the Internet 252, etc. Wait. The remote computer receives the data via the network computer, executes the received commands and transmits the data to the computer system 200 for storage in the storage device. In one embodiment, server 110 is implemented as a computer 200 that includes a power manager 280 that is responsible for managing the server power system. Manager 280 is implemented as a software suite that executes on the server. However, the administrator 280 can execute on other computers connected to the network 100. For convenience of explanation, the power manager 280 is shown in the memory 2〇8 when it is executed. 15 power consumption of the original server 110 components, including cpu 2〇4, memory 2〇8, storage device 216, peripheral interface card, peripheral devices such as printers and external drives, etc. , usually consumes a considerable amount of power. However, each component provides an interface that can be used to power the government. Depending on the component, interface, and manufacturer, the interface may be hardware, firmware, software, hardware that works with the firmware driver and/or software, and so on. These skills in the industry will assume that the firmware can be replaced by other firmware or software programs. In one embodiment, the official 280 controls the components based on the instructions and specifications of the interface to control their power consumption. 1287702 发明, DESCRIPTION OF THE INVENTION The total power consumed by the CPU 204 depends on a number of parameters, such as the frequency at which the processor is executed, the power cycle, the number of processors operating in the system, and the like. The power cycle is the time it takes for the processor to divide by the total time at normal speed. In general, under normal operating conditions, in order to transfer a maximum of 5 computing power, the CPU 204 of the server 110 typically operates at the highest frequency. However, in accordance with the teachings of the present invention, when maximum computational power is not required, the administrator 280 reduces and/or adjusts one or more of the above parameters to reduce the power consumed by the CPU 204 and the server 110. For example, in one embodiment, CPU 204 is implemented as a 10 Pentium 4 manufactured by Intel Corporation of Santa Clara, Calif., which includes a Temperature Control Circuit (TCC) interface. In order to control the speed of the CPU 204 and its power consumption, the manager 280 changes the power cycle of any of the processors 204 through the TCC interface following the specifications and instructions of the CPU. Manager 280 will also set this power cycle to zero at the appropriate time, i.e., effectively stop CPU 204, although CPU 15 is still operational. In another embodiment, the administrator 280 closes the CPU 204 implemented by the Intel processor die and motherboard design through an application that uses the ACPI interface. In one embodiment, RAM 208 operates in slow RAM update and hibernate mode when data is replaced by a disk drive. In order to reduce power consumption, the controller 280 sets the RAM 208 to one of the above two modes because either of the above modes requires less power than the normal mode of operation. The disk drive is implemented as a storage device 216, which also includes an interface for the manager 280 to reduce the speed of the drive to operate in a low power mode. In this mode, the disk stops spinning. 1287702 发明, INSTRUCTION DESCRIPTION In one embodiment, the manager 280 uses the Advanced Assembled Power Interface (ACPI) standard to reduce the power consumption of the server 110. In the 8 〇 state, the server 110 operates normally at full speed. In the S1 state, the manager 28 stops the CPU 204 and updates the RAM 208 to operate the server 110 in the low power mode. In the S2 state, the administrator sets the CPU 204 to ''No power, mode, update RAM 208, and causes the server 11 to operate in a lower power mode than the} state. In the S3 mode, the manager 280 sets the CPU 204 to the ''no power" state, the RAM 208 to the low speed update state, and the server 11's power supply device to the low power mode. In the S4 or hibernation state, the manager 280 10 completely shuts down the hardware and stores the system memory to the disk. In the S5 state, the manager 280 completely shuts down the hardware, shuts down the operating system executed by the server 11, and the server 110 needs to be restarted to return to normal. Mode of Operation. Power Management Based on System Component Load and Application Independent In one embodiment, in order to determine system performance, the administrator 280 does not consider the application components that are executed on the server 11 to measure system load. Usage such as CPU, memory, disk drive, bus, etc. Again, CPU usage is expressed in % (100 * (total time - idle time) / total time); disk and memory I / O Usage is expressed in % (100* (maximum rate - actual rate) / highest rate), and bus usage is expressed in % ((1〇〇* current bus 20 data rate) maximum data rate). For convenience of description, CPU performance is used as an example. However, the concept of the description can also be applied to other components of the server 110 and the performance parameters discussed below. In one embodiment, the manager 280 uses the application provided by the operating system. The program can be programmed, such as Microsoft Windows for 13 1287702, and the Win32 API of the invention system to obtain the usage value. The manager then uses the obtained value to calculate the cpu (4) average and standard using the following formula. Poor: 1 i=n
~X=~nlX' CD 5 (2)~X=~nlX' CD 5 (2)
其中 A =第ith個測量到的CPU使用率值 元=n個測量值之平均cpu使用率 σ =標準差 10 管理者280也以下列公式計算備用容量： 5, = (ΐ00-(χ + ^χσ)) (3)Where A = the ith measured CPU usage value element = the average cpu usage of the n measured values σ = standard deviation 10 The manager 280 also calculates the spare capacity by the following formula: 5, = (ΐ00-(χ + ^ Χσ)) (3)
其中m為可调整之參數，其表示標準差方面的安全限度。 參數m是依據CPU到達飽和的機率或時間百分率作選擇， 例如當cpu到達一預定的使用率值時。在此觀點下，cPU 15或系統效能會降低。為了方便說明，CPU在達到100%使用 率時視為飽和。然而其他接近100%的數值如90、95%等等 亦在本發明的範疇内。 參數m與CPU達到飽和機率之間的關下如下面公式： P(m) = 1 -Where m is an adjustable parameter that represents a safety margin in terms of standard deviation. The parameter m is selected based on the probability or percentage of time the CPU reaches saturation, such as when the cpu reaches a predetermined usage value. In this view, cPU 15 or system performance will be reduced. For ease of explanation, the CPU is considered saturated when it reaches 100% usage. However, other values close to 100% such as 90, 95%, etc. are also within the scope of the present invention. The relationship between the parameter m and the CPU reaching the saturation probability is as follows: P(m) = 1 -
⑷ 20換言之，此關係分別展示於第3A與第3B圖中之表格3〇〇A 與300B 。 14 1287702 玖、發明說明 在一實施例中，管理者280改變伺服器110的運作，例 如以下面公式改變時脈速度或功率週期以獲得所需要之 CPU使用率變動： = -s 七 Η (5) 5 其中//為備用容量想要達到的數值。舉例來說，假若目前 備用容量為10%，而想要將備用容量變動成〇°/。時，管理者 280會將時脈速度降低10%，例如：Ac=-10%+0%=-10%。 然而假如想要將備用容量變動成5%時，Δί-_10%+5%=-5% 。假如想要將備用容量變動成10%時，Ac=-10%+10%=0% 10 等等。換言之，備用容量在一範圍内是可以接受的，例如 從0到Η。為了方便說明，備用容量設定為0%。 一般而言，在時脈速度依據公式（5)調整後，參數m轉 換成CPU在其效能期間會飽和的機率。舉例來說，假若 J =40%、m=2以及π =10時，根據公式（3)可得5=40%，根 15 據公式（5)可得Δ〔=-40%。在此範例中，當m=2時，從公式 (4)或表格300A可得Ρ(2)=2·275。亦即CPU達到飽和的機率 為2.275%。再者，因為Δ(=-40°/〇，CPU時脈速度或功率週 期會降低40%，同時也達到CPU達飽和機率不超過2.275% 的目標。 20 在另一範例中，假若m=3，使用上述公式與表格300可 以得到尸(3)=0.13 5%與Ac=-30%。換言之，在時脈速度降低 30%後CPU有0.135%的機率會達飽和。假若m=0，那麼 尸(0)=0.5以及Δπ·60%，亦即在時脈速度降低60%後CPU有 50%的機率會達飽和。因此較高的m參數數值會導致較高 15 1287702 玖、發明說明 的安全限度’但同時也會降低用以降低CPU速度與電源消 耗的機率。 相反地，如上述說明在CPU速度降低後假若系統付載 增加時，CPU速度會依據公式（5)來增加。舉例來說，假若 5奶=2、P90以及α = 時，根據公式（3)，備用容量p_iQ。 因此，依據公式（5)可得Ac=l〇。時脈速度因而會增加ι〇% 以在所需要的安全縣度下取回所西的電源·效能折衷點。 同樣地，假若m=3、7=90以及^7=10時，則尸_2〇、Δρ2〇 。時脈速度因而增加20%以達到所需要的折衷點。 1〇 在一實施例中，電源管理者280在一週期時間内週期 性地測罝系統資源的使用率值，管理者28〇從這些數值大 致估計或決定具有平均值與變異數之統計分佈。舉例來說 ，對每10分鐘，管理者280每1〇秒鐘取得6〇個使甩率數值 ，並且依據這60個數值計算平均值、標準差、備用容量等 15等。在-實施例中，統計分佈是以標準或高斯分佈取得。 擷取使用率數值的週期一般稱之為取樣週期，取樣點&會 隨許多因素變動，這些因素包括系統使用率是否穩定或波 動、也會取決於每天的時間、每週的每一天等等。舉例來 說，在白天期間因為不同的工作客戶請求不同的服務而使 2〇系統負載可能有較大的波動。因此，在此期間的取樣週期 會比系統使料較穩定的晚上或週末來得短。同樣地，工 作日的取樣數目會高於晚上或週末的取樣數目。 一般而言，參數所的數值是依據描述系統資源如cpu 達到飽和可接受的機率之測量資料與所需要安全限度的機 16 1287702 玖、發明說明 率分佈作選擇。在一實施例中，此嬙査 此機率以及/或參數m與服 務層級協；致。假如服務應用程式是關鍵性的，因此必 需盡可能的避免系統飽和時，則會選擇高的參數_如5。 然而假如此應用程式並非關鍵性的，那麼⑺可以選擇為 如同上面說明，當w等於0時’在CPU時脈速率依據公式 (5)調整後CPU將有50%的機率達到飽和。 10 15(4) 20 In other words, this relationship is shown in Tables 3A and 300B in Figures 3A and 3B, respectively. 14 1287702 发明, DESCRIPTION OF THE INVENTION In one embodiment, the manager 280 changes the operation of the server 110, such as changing the clock speed or power cycle with the following formula to obtain the required CPU usage change: = -s 七Η (5 ) 5 where / / is the value that the spare capacity wants to reach. For example, if the current spare capacity is 10%, you want to change the spare capacity to 〇°/. At the time, the manager 280 will reduce the clock speed by 10%, for example: Ac = -10% + 0% = -10%. However, if you want to change the spare capacity to 5%, Δί-_10%+5%=-5%. If you want to change the spare capacity to 10%, Ac = -10% + 10% = 0% 10 and so on. In other words, the spare capacity is acceptable over a range, for example from 0 to Η. For convenience of explanation, the spare capacity is set to 0%. In general, after the clock speed is adjusted according to equation (5), the parameter m is converted to the probability that the CPU will saturate during its performance. For example, if J = 40%, m = 2, and π = 10, 5 = 40% can be obtained according to formula (3), and root 15 can obtain Δ [= -40% according to formula (5). In this example, when m = 2, Ρ (2) = 2.275 can be obtained from equation (4) or table 300A. That is, the probability of the CPU reaching saturation is 2.275%. Furthermore, because Δ(=-40°/〇, the CPU clock speed or power cycle is reduced by 40%, and the CPU achieves a saturation probability of no more than 2.275%. 20 In another example, if m=3 Using the above formula and table 300, you can get the corpse (3)=0.13 5% and Ac=-30%. In other words, the CPU has a probability of saturation of 0.135% after the clock speed is reduced by 30%. If m=0, then The corpse (0) = 0.5 and Δπ · 60%, that is, the CPU has a 50% chance of saturation after the clock speed is reduced by 60%. Therefore, a higher m parameter value will result in a higher 15 1287702 玖, the invention description The safety margin 'but also reduces the probability of reducing CPU speed and power consumption. Conversely, as described above, if the system load increases after the CPU speed decreases, the CPU speed will increase according to formula (5). Say, if 5 milk = 2, P90 and α =, according to formula (3), spare capacity p_iQ. Therefore, according to formula (5) can get Ac = l 〇. The clock speed will increase ι 〇 % Retrieve the power and performance compromise points of the West under the required security county. Similarly, if m=3, 7=90, and ^7=10 Then, the corpus 2 〇, Δρ2 〇. The clock speed is thus increased by 20% to reach the required compromise point. In an embodiment, the power manager 280 periodically measures system resources during a cycle time. The usage rate, the manager 28 大致 roughly estimates or determines the statistical distribution with the mean and the variance from these values. For example, for every 10 minutes, the manager 280 obtains 6 甩 rate values every 1 〇 second. And calculating the average value, standard deviation, spare capacity, etc. according to the 60 values. In the embodiment, the statistical distribution is obtained by a standard or Gaussian distribution. The period of the usage rate value is generally referred to as a sampling period. Sampling points &amps vary with many factors, including whether the system usage is stable or fluctuating, depending on the time of day, every day of the week, etc. For example, during the day due to different work customer requests Different services may cause the system load to fluctuate greatly. Therefore, the sampling period during this period will be shorter than the night or weekend when the system is stable. The number of samples taken on a day will be higher than the number of samples taken at night or on weekends. In general, the value of the parameter is based on the measurement data describing the system resource such as cpu reaching saturation acceptable probability and the required safety limit 16 1287702 发明, invention Explain the rate distribution as a choice. In an embodiment, this check rate and/or parameter m is related to the service level; if the service application is critical, it is necessary to avoid system saturation as much as possible. Choose a high parameter _ such as 5. However, if the application is not critical, then (7) can be selected as explained above, when w is equal to 0, 'the CPU will have 50% after the CPU clock rate is adjusted according to formula (5). The chances are saturated. 10 15
在一實施例中，在對每-資源的安全限度給-合適的 祕，管理者，將與資源對應的備用容量s設定至一預定 的範圍，此範圍實際上盡可能地接近零。如此可將系統資 源用的系統電源最佳化。依據這些資源，會調整不同的變 數以調整備用容量。舉例來說，在單_處理器的狀況下， 會根據上述說明調整與處理器使料對應的時脈速度。在 多處理器的情況下，可以開啟或關閉—或多個處理器。舉 例來說，假若是使用P4處理器，當備用容量為5G或更高時In an embodiment, the appropriate capacity, for the security limit for each resource, is set by the administrator to set the spare capacity s corresponding to the resource to a predetermined range, which is actually as close as possible to zero. This optimizes the system power for system resources. Based on these resources, different variables are adjusted to adjust the spare capacity. For example, in the case of a single-processor, the clock speed corresponding to the processor material is adjusted according to the above description. In the case of multiple processors, you can turn on or off—or multiple processors. For example, if you are using a P4 processor, when the spare capacity is 5G or higher.
，則會關閉2個處理n崎低備用容量。假若資源為磁碟 機時，則調整旋轉速度等等。 包含表格300A之第3A圖展示參數所與(：1>1；使用率達到 100%之機率户(m)之間的第一種關係表。在表格3〇〇a中， 再給定數值m時可以取得機率P(w)。舉例來說，對於數值 20 W=_2·99，選擇行 _3·0 與列-〇·〇1 則可得 m=-3.0+0.01=-2.99 以 及Ρ=99·865。換言之，Ρ(-2 99)=99 865。在此範例中對於 數值m=-2.99而言，cpu使用率有99 865%的機率會達1〇〇% 。同樣地’當所=1.01時，選擇行1與列〇]可得所=1+〇」以 及户=15.625 ;或者/>(1·01)=15·625。因此對數值^ 〇1而 17 1287702 玖、發明說明 言，CPU使用率有15 625%的機率會達1〇〇%。 包含表袼300B第3B圖展示參數所與CPU使用率達到 100/〇之機率户(⑷之間的第二種關係表。在表袼刈⑽中， 再給定機率户時可以取得參數值所。舉例來說，當機率為 5 49%時，選擇行48與列1可的^=48 + 1=49以及w=0.0251。在 此範例中，假若CPU使用率為i 〇〇%之可接受的時間為總時 間的49%時，那麼於公式（3)中會選擇所=〇〇25。同樣地， 當機率為2·8%時，選擇行2與列〇·8可得尸=2+〇 8==2 8以及 痛 所=1·911。因此假若可以接受有28%的時間cpu使用率為 10 1〇0%時，那麼於公式（3)中會選擇m=1.911。在業界這些技 · 藝會被認為表格300A和300B與上述公式（4)為不同的表示 方式。 依據應用私式效能或服務層級協定之電源管理 除了系統資源使用率外，管理者28〇依據伺服器11〇上 15執行的應用程式以及或由服務層級協定（SLA)定義之效能 參數來測量系統效能。這些參數包括反應時間、使用者身 · 份驗證數量、或是每一單位時間提供的網頁數量等等。反 應時間可以有個種不同的定義，例如從使用者請求一服務 到使用者接收到反應之間的時間、應用程式用以服務一請 20求所花費的時間等等。管理者280經由應用程式界面（Αρι) · 存取效能參數。在一實施例中，管理者28〇藉由呼叫適當 . 的API來測量這些參數並依據服務層級協定比較這些參數 以決定哪些是需要的，在一實施例中，此服務層級協定代 表一臨界值。假若伺服器110輸出的效能高於臨界值時， 18 1287702 玖、發明說明 e理者280藉由上述一或多個機構的組合來降低電源消耗 。亦即官理者280降低時脈速度、關閉某些處理器、降低 磁碟機旋轉速度等等。同樣地，當伺服器丨1〇無法依據 SLA來執行時，官理者28〇增加系統電源以使伺服器Η。可 5以增加其效能進而符合ALS之要求。 ‘ 一般而言，一或多個資源使用率參數的組合會影響 1 SLA參數。舉例來說，cpu使用率、磁碟存取時間等等中 的一個或多個組合會影響檢索網頁的反應時間、每分鐘驗 · 證使用者身份的數量等等。在一實施例中，為了符合sla 1〇規定，管理者28〇會考慮數對參數之間的關係。為丁方便 · 扰明，使用分別對應CPU平均使用率與服務網頁平均反應 時間之一對變數1與7。管理者28〇決定變數尤與變數少之間 的關係。在-實施例中，管理者28〇使用如位於加州佛利 蒙之F〇rtel公司製作之SightUne工具來取得相關係數仏， b其表示變數X與變數少之間的相關度。係數々=1表示咖是 相關的，亦即當X改變部分量時，少也會以相同的量改變。 φ 舉例來況，當X改變1〇%時，少也會改變1〇%。當讀變鳩 時，少也會改變贏等等。相反地，假如产〇時，χ與戍各 自獨立的’因此改變其中_個參數並不會影響另—個參數 · 20 。然而當々介於咖之間時，讀少之間有某程度的相關性 ·· 。管理者280使用下列公式⑹來決定咖雙變數分佈之機 · 率分佈函數： 19 1287702 玖、發明說明 f{x^y)= 1 --j==QXp< 2π^χ^γψ-ρ2 其中 X =第一個變數的平均值，例如CPU平均使用率 无=平均值X之平均值 5 ^ =第二個變數的平均值，例如服務一網頁之平均 反應時間 歹=平均值少之平均值 a = X之標準差 ~ 之標準差 10 — 1與>;之間的相關係數 為了確定公式（6)的函數/^，少），管理者28〇週期性地測 里數個對應的反應時間與cpu使用率，並對每一量測到的 數值組計算其對應的平均值。同樣地，管理者28〇從數個 數值組或數個平均值中計算這些平均數值之平均值與對靡 15 的標準差。 、… 函數/(x，>〇可以以包括座標軸χ、少與/(χ，少)之3維圖形表 示，其中X座標與;;座標構成^少平面。從公式（6)，可以取 得表示變數X與變數y相互關係之趨勢線。此趨勢線為 最大值投影至Xy平面之執跡，其最大值是使用^^ = 〇八 dy A 2〇式求得。根據數學計算，此趨勢線可以表示成下面公式： Y=pJ^x-pJ-^x,Y (7) 一 1 2y^7) χ-χ _2Ac，j jc-x σγ y~y y-y — 一 σ 1287702 玫、發明說明 其斜率為： P = Px,y ~ (8) 第4圖展示一趨勢線400，其中X軸表示CPU平均使用 率，7軸以秒為單位表示平均反應時間。在趨勢線400上，, will close 2 processing n low standby capacity. If the resource is a disk drive, adjust the rotation speed and so on. Figure 3A, which contains table 300A, shows the first relationship between the parameters and (:1>1; probability (100) usage rate. In Table 3〇〇a, the value m is given. The probability P(w) can be obtained. For example, for the value 20 W=_2·99, select row _3·0 and column 〇·〇1 to get m=-3.0+0.01=-2.99 and Ρ= 99.865. In other words, Ρ(-2 99)=99 865. In this example, for the value m=-2.99, the probability of cpu usage is 99 865%, which will reach 1〇〇%. =1.01, select row 1 and column 〇] to get =1+〇” and household=1.625; or />(1·01)=15·625. Therefore, the value ^ 〇1 and 17 1287702 玖, invention Explain that the CPU usage rate of 15 625% will reach 1%. Contains Table 300B Figure 3B shows the second relationship between the parameters and the probability that the CPU usage reaches 100/〇 ((4) In Table (10), the parameter value can be obtained when the probability is given. For example, when the probability is 5 49%, select row 48 and column 1 can be ^=48 + 1=49 and w= 0.0251. In this example, if the CPU usage is i 〇〇% acceptable When the time is 49% of the total time, then in the formula (3), the ==25 is selected. Similarly, when the probability is 2.8%, the row 2 and the column 〇8 can be selected to obtain the corpse = 2+ 〇8==2 8 and pain station=1·911. Therefore, if it is acceptable to have 28% of the time cpu usage rate is 10 1〇0%, then in formula (3), m=1.911 will be selected. The technique is considered to be a different representation of the tables 300A and 300B from the above formula (4). The power management according to the application of the private performance or the service level agreement, in addition to the system resource usage rate, the manager 28 depends on the server 11〇 System performance is measured by the application executed on the top 15 and by the performance parameters defined by the service level agreement (SLA). These parameters include response time, number of user authentications, or number of pages provided per unit time, etc. There may be different definitions of the reaction time, such as the time between when the user requests a service and when the user receives the response, the time the application spends servicing the request, etc. The manager 280 passes the application. Program interface (Αρι) · Save Performance parameters. In one embodiment, the manager 28 measures these parameters by calling the appropriate API and compares the parameters according to service level agreements to determine which ones are needed. In one embodiment, the service level agreement represents A threshold value. If the output of the server 110 is higher than the threshold, 18 1287702 发明, the invention 280 is reduced by the combination of one or more of the above mechanisms to reduce power consumption. That is, the official 280 reduces the clock speed, turns off certain processors, reduces the speed of the disk drive, and the like. Similarly, when the server 〇1〇 cannot be executed in accordance with the SLA, the official 28 increases the power of the system to cause the server to falter. Can be 5 to increase its efficiency and thus meet the requirements of ALS. ‘ In general, a combination of one or more resource usage parameters affects 1 SLA parameter. For example, one or more combinations of cpu usage, disk access time, and the like can affect the response time of the retrieved web page, the number of authenticated users per minute, and the like. In an embodiment, the manager 28 will consider the relationship between pairs of parameters in order to comply with the sla 1 〇 specification. For the convenience of Ding · Disturbing, use the corresponding average CPU usage and the average response time of the service page to the variables 1 and 7. Manager 28 determines the relationship between variables and variables. In the embodiment, the manager 28 uses the SightUne tool, such as that produced by F〇rtel, Inc. of Fremont, Calif., to obtain the correlation coefficient 仏, b which represents the correlation between the variable X and the variable. A coefficient 々 = 1 indicates that the coffee is related, that is, when X changes the partial amount, less is changed by the same amount. φ For example, when X changes by 1〇%, it will change by 1%. When reading changes, less will change the win and so on. Conversely, if the calving occurs, the χ and 戍 are each independent' so changing the _ parameter does not affect the other parameter. However, when there is a gap between the coffee and the coffee, there is a certain degree of correlation between the readings. The manager 280 uses the following formula (6) to determine the machine-rate distribution function of the coffee double variable distribution: 19 1287702 玖, invention description f{x^y) = 1 --j==QXp< 2π^χ^γψ-ρ2 where X = the average of the first variable, such as the average CPU usage = the average of the average X 5 ^ = the average of the second variable, such as the average response time of a service page 歹 = the average of the average less a = standard deviation of X ~ the standard deviation of 10 - 1 and >; in order to determine the function /^ of formula (6), the manager 28 〇 periodically measures the corresponding reaction time Calculate the corresponding average value with the cpu usage rate and the value group measured for each quantity. Similarly, the manager 28 calculates the average of these average values and the standard deviation of 靡 15 from a number of numerical groups or a plurality of average values. The function /(x,>〇 can be represented by a 3-dimensional graph including the coordinate axis 少, less and / (χ, less), where the X coordinate and ;; coordinates constitute a small plane. From equation (6), can be obtained A trend line representing the relationship between the variable X and the variable y. This trend line is the maximum projection to the Xy plane, and its maximum value is obtained using ^^ = 〇8 dy A 2〇. According to mathematical calculations, this trend The line can be expressed as the following formula: Y=pJ^x-pJ-^x, Y (7) -1 2y^7) χ-χ _2Ac,j jc-x σγ y~y yy — σ 1287702 玫,发明说明The slope is: P = Px, y ~ (8) Figure 4 shows a trend line 400 where the X-axis represents the average CPU usage and the 7-axis represents the average response time in seconds. On the trend line 400,
5 t/點表示有50%的時間當CPU使用率為80%時反應時間為一 秒或更短；F點表示有50%的時間當CPU使用率為100%時 反應時間為1.25秒或更短等等。5 t / dot means 50% of the time when the CPU usage is 80%, the reaction time is one second or less; F point means 50% of the time, when the CPU usage is 100%, the reaction time is 1.25 seconds or more. Short and so on.
管理者2 8 0根據服務層級協定對變數:μ產生所需要的數 值D以及此所需數值Ζ)之信心度。此信心度表示變數y等於 10 或低於數值Z)之機率。為了方便說明，Z)為一秒鐘，假若 信心度為60%、70%、80%時，平均反應時間等於或低於 一秒鐘的機率分別為60%、70%、80%。這些技術在業界 會認為假若信心度為50時，數值D是對應至趨勢線400上的 C/點。再者，X軸上對應的CPU使用率為80%或數值£：。然 15 而，假若想要達到95%時間之數值Z)時，管理者280在少軸 上選擇比一秒更快的目標反應時間Γ。此一目標數值Γ提供 一限度使得95%的時間，而非50%的時間其反應時間低於 數值D之一秒鐘。在一實施例中，管理者280使用變數m來 決定機率值95%，在表格300B中顯示m等於1.6449。變數m 20 也是用以在低於數值下達到數值Γ之標準差數值。數值Γ與 數值D之間的關係為 T = D-{mxay) (9) 管理者280使用下列公式來決定備用容量，以變數少之 21 (10) 1287702 玖、發明說明 標準差之數值m表示： 其中 \二備用容量 5 ^"^之目標值 Α =少之實際值 S =3^之標準差 w一於所需要信度對應之標準差數值 ® 在公式（10)中，數值絲示變數少之實際或實際時間數 - 10值’其通常在計算公式1〇)之前一刻才取得。在變數少為平 均反應時間的實施例中，數值後依據週期性量測之數個 反應時間值來決定。 管理者280會依據數值Γ來決定對應的cpu使用率及， 其表示當CPU以Λ值運作時，有95%的信心度表示反應時間 15為數值D或更短，或者以上述的範例來說為一秒鐘。 · 使用趨勢線400，與從數值D變化至數值Γ對應之變數X 從數值五變動至數值i?之變化Ax之計算如下： Δχ = ^- Ρ · 其中/7為趨勢線400之斜率，定義於上述公式（8)中。口工可 20 以視為變數X之備用容量、或者視為CPU使用率之備用容 量。 在一實施例中，管理者280改變伺服器11〇的運作，例 22 玖、發明說明 如改變所需要的時脈速度或 需要的改變： 、題期以獲得CPU使用率所The manager 2800 determines the confidence of the variable D and the required value Ζ) according to the service level agreement. This confidence indicates the probability that the variable y is equal to 10 or lower than the value Z). For convenience of explanation, Z) is one second. If the confidence is 60%, 70%, or 80%, the average reaction time is equal to or lower than one second, and the probability is 60%, 70%, and 80%, respectively. These techniques are considered in the industry to assume that if the confidence level is 50, the value D corresponds to the C/point on the trend line 400. Furthermore, the corresponding CPU usage on the X-axis is 80% or the value £:. However, if it is desired to achieve the value Z) of 95% of the time, the manager 280 selects a target reaction time that is faster than one second on a few axes. This target value Γ provides a limit such that 95% of the time, rather than 50% of the time, is less than one second of the value D. In one embodiment, manager 280 uses the variable m to determine the probability value of 95%, and in table 300B, m is equal to 1.6449. The variable m 20 is also used to reach the standard deviation value of the value 低于 below the value. The relationship between the value Γ and the value D is T = D-{mxay) (9) The manager 280 uses the following formula to determine the spare capacity, with a variable number of 21 (10) 1287702 玖, the value of the standard deviation of the invention indicates m : where \ 2 spare capacity 5 ^"^ target value Α = less actual value S = 3^ standard deviation w - the standard deviation value corresponding to the required reliability ® In formula (10), the numerical value shows The actual or actual number of variables with a small number - 10 values 'which is usually obtained before the calculation formula 1 〇). In the embodiment where the number of variables is less than the average reaction time, the value is determined based on the number of reaction time values measured periodically. The manager 280 will determine the corresponding cpu usage rate according to the value 及, which means that when the CPU operates at the Λ value, 95% confidence indicates that the reaction time 15 is the value D or shorter, or in the above example For one second. · Using the trend line 400, the change from the value D to the value Γ from the value five to the value i? The change Ax is calculated as follows: Δχ = ^- Ρ · where /7 is the slope of the trend line 400, defined In the above formula (8). Oral workers can be considered as spare capacity for variable X or as an alternate capacity for CPU usage. In one embodiment, the manager 280 changes the operation of the server 11 ,, example 22, invention instructions such as changing the required clock speed or the required changes: , the title period to obtain CPU usage
Δ x+GΔ x+G
Wg為if用容量所需要達 前的備用容量為10%並且相 值。舉例來說’假若目 理者280會料脈速度較1G%、’”容量«為零時，管 然而假若想要將備用容量變動如△ r·舰+G%=-10%。 。然而假若想要將備用容量變動為^，則Δ"=_1()%+5%=-5% 10 等等。換言之，備用容量在—1Γ時，則—％+3%=-7% 例如從〇至數值G。 &乾圍内時是可以接受的’ &理者280也會經由_或吝 雷、店 X夕個不同計數之組合來節省 電源，例如最佳化效能、結合 孜冲 由不同伺服器110執行之服 力、付合財的條件、關閉不重要或非關鍵性的程式等等。 效能最佳化 因為效能最佳化可以在給定的負載下創造更多的系統 備用谷里，電源官理者280使用效能最佳化來降低電源消 耗以維持相同的系統效能。在一實施例中，為了週期性地 最佳化系統效能，系統監控者（隨作業環境而有所不同）使 用數種可取付的工具如位於加州Alto之HewlettPackard公 20 司製作之OpenView、GlancePlus、GlancePlus Pak 2000、 位於紐約 Islandia之 Computer Associates International 製作 給開放VMS用之Unicenter Performance Management、位於 紐約 Armonk 之 IBM 公司製作之 Tivoli ApplicationThe spare capacity required for Wg to be the if capacity is 10% and the phase value. For example, if the 280 is expected to have a pulse rate of 1 G% and 'Capacity' is zero, if you want to change the reserve capacity as Δr·ship+G%=-10%. If you want to change the spare capacity to ^, then Δ"=_1()%+5%=-5% 10 etc. In other words, when the spare capacity is at -1Γ, then -%+3%=-7%, for example, from To the value G. & is acceptable in the dry square. '& 280 will also save power by a combination of _ or 吝雷, shop X 不同 different counts, such as optimization performance, combined with the hedge Different server 110 performs the service, balances the conditions, closes the unimportant or non-critical programs, etc. Performance optimization Because performance optimization can create more system backup valleys under a given load. Power registrar 280 uses performance optimization to reduce power consumption to maintain the same system performance. In one embodiment, system monitors (depending on the operating environment) in order to periodically optimize system performance Use several affordable tools such as OpenVi from Hewlett Packard, Inc., Alto, Calif. Ew, GlancePlus, GlancePlus Pak 2000, Computer Associates International, Islandia, New York, produced by Unicenter Performance Management for Open VMS, Tivoli Application for IBM, Armonk, NY
Performance Management等等。一般而言，一旦經過效能 23 1287702 玖、發明說明 最佳化後，管理者280使用一種節省電源技術來管理系統 電源。 在一貫施例中，效能最佳化是藉由調整各種動態或靜 態“可調整，，的參數達成。動態參數可以在正常的系統運作 · 5期間麦動而不需要重新啟動。動態參數包括伺服器110上 - 所執行之所有應用程式與服務的優先權層級、快取網頁與 . 資料庫中間查詢結果之應用層級記憶體快取大小、同時存 取服務或登入至一服務之使用者人數、應用程式用之CPU · 時序分割限制等等。靜態參數需要重新啟動系統以產生作 10用，包括的參數如磁碟機陣列（raid)組配、置換檔大小、 系統記憶體大小、系統快取大小等等。 群組與叢集式系統 在一實施例中，當一組伺服器110是在叢集或負載平 衡％境下運作時，管理者280依據每一伺服器11〇的效能在 15適當的時機將應用程式合併至較少的系統。舉例來說，假 若兩台伺服器110-1與110—2提供網頁服務，並且當此二伺 · 服器的總備用容量超過1〇〇%時，將伺服器11〇_丨設定至較 高的‘‘睡眠，，狀態如4或85狀態甚至關機，而伺服器11〇2執 行自己本身與伺服器HOd的應用程式。另一範例：執行 ‘ 20資料庫應用程式之伺服器no-3的備用容量為20%，沒有執 ·_ 行資料庫應用程式時備用容量增加至5〇0/〇。在一實施例中 · ，會關閉在伺服器110-3執行的資料庫應用程式或是轉移 至伺服器110-4，因此伺服器110-3的5〇%容量可以用以提 供伺服HUG·執行時所需之5G%容量。在_實施例中，# 24 1287702 玖、發明說明 2器的備用容量為CPU之備用容量。在許多情況下，服務 I合即省之電源高於分別降低每台伺服器中某些資源之電 源。 預設狀況 5 官理者280也會依據預設的狀況與需求來執行電源管 理。在特殊或例外的狀況下，管理者28〇依據事先提供的 組配資訊在適當的時機關閉在主要環境條件下並非絕對必 要或不重要的％式或應用程式。關閉不重要的程式或非關 鍵性的服務會增加其他服務的效能進能得到額外的容量。 10特殊條件是由外部事件觸發，例如“電壓不足，，狀況、超過 電流額度時間、煙霧與“空氣濾清，，警告、因為天氣炎熱造 成部分設備空調過載等等。 管理系統電源方法之步驟說明 第5 A圖為說明依據一實施例管理系統電源方法之流程 15 圖。 在步驟502中，管理者280週期性地將系統效能最佳化。 在步驟504中，管理者280在η個測量值中取得各種系 統為源之使用率數值t。 在步驟506中，管理者280對對應的資源計算X/之平均 20 值f與標準差π。 在步驟508中，管理者280取得可接受之資源達到飽和 的機率。亦即管理者280取得公式（4)之機率Ρ。 在步驟512中，管理者280從機率Ρ中取得參數所。 在步驟516中，管理者280從取得的f、π、m數值計 25 1287702 5久、發明說明 算備用容量s。 假若步驟5 18判定備用容量在可接受的範圍内時，表 示系統的運作是可接受的，此方法接著回到步驟502。 然而假若步驟5 18指示出備用容量並非在可接受的範 5 圍内時，管理者280在步驟520中計算資源使用率的變化率 Δ c。 在步驟524中，為了反應之變動，管理者28〇調整 系統之運作，例如CPU時脈速度。一旦資源使用率被調整 後’備用容量s應該會在可接受的範圍内。此方法接著回 10 到步驟502並繼續電源管理程序。 · 第5B圖為說明依據一實施例管理與SLA參數有關之系 統電源方法之流程圖。 在步驟526中，管理者280週期性地將系統效能最佳化。 在步驟528中，管理者280對反應時間與對應的(：1>1;使 15用率作取樣。管理者280從每一取樣叢集取得對應的平均 值，管理者280從數個取樣叢集取得這些平均數的平均值 · 以及其對應的標準差。 在步驟530中，管理者28〇決定cpu使用率與反應時間 之間的相關係數。 ^ W 在步驟534中，管理者280產生一關係圖，例如反應時 ·* 間與CPU使用率之趨勢線。 . 在步驟538中，管理者28〇依據服務層級協定產生所需 要反應時間之數值Z)，例如一秒鐘，以及與此數值㈣應 之信心度。假如信心度為5〇%時，可以直接從公式⑺中得 26 1287702 玖、發明說明 到與此數值Z)對應之CPU使用率。然而，為了方便說明， 此信心度高於50%，例如95°/。。 在步驟542中，管理者280決定比一秒鐘更快之目標反 應時間值，例如數值Γ，使得當反應時間為數值7時，反應 5 時間有95%的機率會低於數值D。 在步驟546中，管理者280依據目標值Γ計算反應時間 之備用容量。 在步驟550中，管理者280計算與反應時間備用容量對 應之CPU備用容量。 10 假若步驟554判定CPU的備用容量在可接受的範圍内 時，表示系統的運作是可接受的，此方法接著回到步驟 502。然而假若步驟554指示出備用容量並非在可接受的範 圍内時，管理者280在步驟558中計算CPU使用率的變化率。 在步驟562中，管理者280調整伺服器110之運作，包 15 括調整CPU時脈速度使得CPU備用容量在可接受的層級。 此方法接著回到步驟502並繼續電源管理程序。 在第5A與5B圖之上述步驟期間，管理者可以可以在 符合某些預設的狀況時關閉某些應用程式、合併應用程式 等等。 20 電源管理者 第6圖展示包含管理程式610、效能測量程式620、效 能控制程式630、效能最佳化程式640以及程式腳本650與 660，全部經由匯流排6050通訊之電源管理者280實施例。 在一實施例中，程式610是以Windows服務或Unix程序 1287702 玖、發明說明 方式執行’並且包含一管理使用者界面6150與啟動程式 6160。使用者界面6150允許使用者輸入組配資訊、執行指 令、腳本以及/或在圖形使用者界面（GUI)中作選擇。啟動 程式6160啟動效能測量程式620、效能控制程式63〇以及啟 5 動時之腳本’並且處理初始化工作，例如經由界面615 〇讀 取先前輸入並儲存之組配。 效能測量程式620監測系統資源效能與包含在服務層 級協定之參數。效能程式620計算效能參數之統計平均值 與標準差，並且依據服務層級協定比較何者是必須的。程 10式620也會計算備用容量值。程式62〇經由適當的元件與應 用程式界面不間斷地測量並記錄這些參數，並且當有需要 時，觸發控制程式630作適當的動作來管理電源。 效能控制程式630使用一種節省電源方式來降低或增 加電源。控制程式630包括控制系統資源所需要之資％與 15 界面。舉例來說，控制程式630經由TTC電路降低或增加 CPU之有效時脈速度、增加或降低磁碟機轉速或是在必要 時開啟或關閉上述裝置。 在一實施例中，效能最佳化程式640對系統效能與“可 調整”參數間之關係建立動態模型。程式64〇根據此模型調 20 整這些參數以最佳化系統效能，如上面所討論的，一旦系 統最佳化，測量程式620與控制程式63〇會在適當時機應用 一種上述討論的各種節省電源技術。 腳本650與660依據主要狀況與需求執行電源管理策略 。在一實施例中，腳本650在正常運作時執行以實現收集 28 1287702 玖、發明說明 與分析效能資訊之邏輯。舉例來說，腳本65〇觸發效能測 里程式620计异效能參數之統計平均值與標準差、測量備 用容量等等。腳本650根據測量的結果觸發控制程式63〇以 控制電源。 5 腳本660提供處理例外運作狀況之邏輯。腳本660依據 組配資訊關閉或啟動對應的程式。舉例來說，當遇到“電 力不足’’的狀況時，腳本660觸發控制程式630用以降低電 源。腳本660亦會關閉不重要的程式等等。 執行電源管理者之說明 10 第7圖為說明依據一實施例執行電源管理方法之流程 圖。 在步驟704中，系統管理者在伺服器11〇上安裝電源管 理者。 在步驟708中，管理者經由界面6150組配各種資訊， 15 例如每天在伺服器110上執行程式與服務的時間。管理者 指定非必須但若狀況允許時想要執行的程式、非必要執行 但會依需求執行的程式等等。管理者規劃應用程式可以關 閉、具有較低優先權或依需要而可以執行等等之時間。管 理者將腳本650規劃程正常狀況，並將腳本660規劃成例外 20 狀況。管理者規劃當符合狀況時執行腳本。管理者可以依 需要在步驟708經常變更此資訊與規劃。 在步驟712中，在系統開機後，程式61〇執行正常運作 用之腳本650。程式610也會啟動程式620與程式630以測量 及最佳化系統效能。 29 1287702 玫、發明說明 在步驟716中當外在環境條件變動時，管理者使用界 面6150來發送指令通知電源管者28〇這些更動。電源管理 者280依據這些新的狀況經由程式61〇執行對這些新狀況組 配之腳本650與660。 在上述詳細解說中，本發明是經由參考特定示範實施 例來做解說。然而可以有各種的修改與更動而不背離本發 之主要精神與範疇。因此，詳細解說與圖示只是用以說明 用而非其限制。 【圖式簡單說明】 第1圖展示根據本發明實施例可能實現的網路。 第2圖展示根據本發明實施例可能實現的示範電腦。 第3 A圖展示在給定參數所時用以找出機率之對照表。 第3B圖展示在給定機率p時用以找出參數所之對照表。 第4圖展示兩個變數尤與少之走勢曲線 第5A圖為解說依據一實施例管理系統電源步驟之流程 圖。 第5B圖為解說依據一實施例管理與SLA參數相關之系 統電源步驟之流程圖。 第6圖展示依據一實施例之電源管理。 第7圖為解說依據一實施例執行電源管理之流程圖。 30 1287702 玖、發明說明 【圖式之主要元件代表符號表】 100···網路 252…全球資訊網 110···伺服器 256···伺服器 120-1，102-2，102-N …客戶端 150···通訊鏈結 200···電腦糸統 204···中央處理單元(CPU) 208…隨機存取記憶體(RAM) 212···唯讀記憶體(ROM) 216···儲存裝置 220···通訊界面 224···匯流排 228···輸入裝置 232···顯示裝置 236···游標控制 240…網路 248…印表機 280···管理者 400…趨勢線 502,504,506,508,512,516,518， 520,524,526,528,530,534,538, 542,546,550,554,558,562"涉驟 610···管理程式 620···效能測量程式 630···效能控制程式 640…效能最佳化程式 650,660···程式腳本 704,708,712,716〜步驟 6050…匯流排 6150…使用者界面 6160…啟動程式Performance Management and more. In general, once the performance specification 23 1287702 is optimized, the administrator 280 uses a power saving technique to manage the system power. In a consistent application, performance optimization is achieved by adjusting various dynamic or static "adjustable," parameters. Dynamic parameters can be used during normal system operation. No need to restart. Dynamic parameters include servo. On the device 110 - the priority level of all applications and services executed, the cache page and the application level memory cache size of the database query result, the number of concurrent access services or the number of users logging into a service, CPU for application · Timing split limit, etc. Static parameters need to restart the system to generate 10 parameters, including parameters such as disk array (raid), replacement file size, system memory size, system cache Size and the like. Group and Cluster System In one embodiment, when a group of servers 110 are operating in a cluster or load balancing context, the administrator 280 is appropriate at 15 according to the performance of each server 11 Timing merges the application into fewer systems. For example, if two servers 110-1 and 110-2 provide web services, and when the two servers are total When the capacity exceeds 1〇〇%, the server 11〇_丨 is set to a higher ''sleep, the state is 4 or 85 state or even shut down, and the server 11〇2 executes its own application with the server HOd. Another example: the spare capacity of the server no-3 executing the '20 database application is 20%, and the spare capacity is increased to 5〇0/〇 when there is no database application. In an embodiment If the database application executed by the server 110-3 is closed or transferred to the server 110-4, the capacity of the server 110-3 can be used to provide the servo HUG. 5G% capacity. In the embodiment, #24 1287702 玖, invention description 2 spare capacity is the spare capacity of the CPU. In many cases, the service I is the same as the power supply is lower than each of the servers The power of these resources. Preset status 5 The official 280 will also perform power management according to the preset conditions and requirements. Under special or exceptional conditions, the administrator 28 will provide the appropriate timing based on the information provided in advance. Closing is not absolute under major environmental conditions If it is not important, or if it is not important, turning off unimportant programs or non-critical services will increase the performance of other services to get extra capacity. 10 Special conditions are triggered by external events, such as "voltage shortage, Condition, excess current limit time, smoke and "air filtration, warning, partial air conditioning overload due to hot weather, etc. Steps to manage the system power supply method. Figure 5A is a diagram illustrating the method of managing the system power according to an embodiment. Flowchart 15 Figure 280. The manager 280 periodically optimizes system performance in step 502. In step 504, the manager 280 retrieves usage values t for various systems as sources from the n measurements. In step 506, the manager 280 calculates an average 20 value f of the X/ and a standard deviation π for the corresponding resource. In step 508, the manager 280 obtains the probability that the acceptable resources are saturated. That is, the manager 280 obtains the probability of the formula (4). In step 512, the manager 280 retrieves the parameters from the probability Ρ. In step 516, the manager 280 calculates the spare capacity s from the obtained f, π, and m numerical values 25 1287702 for a long time. If step 5 18 determines that the spare capacity is within an acceptable range, indicating that the operation of the system is acceptable, the method then returns to step 502. However, if step 5 18 indicates that the spare capacity is not within an acceptable range, the manager 280 calculates the rate of change Δ c of the resource usage rate in step 520. In step 524, the manager 28 adjusts the operation of the system, such as the CPU clock speed, in response to changes in the response. Once the resource usage is adjusted, the spare capacity s should be within an acceptable range. This method then returns 10 to step 502 and continues the power management process. Figure 5B is a flow diagram illustrating a method of managing system power associated with SLA parameters in accordance with an embodiment. In step 526, the manager 280 periodically optimizes system performance. In step 528, the manager 280 samples the response time and the corresponding (:1>1; 15 usage rate. The manager 280 obtains the corresponding average from each sample cluster, and the manager 280 obtains from the plurality of sample clusters. The average of these averages and their corresponding standard deviations. In step 530, the manager 28 determines the correlation coefficient between the cpu usage rate and the reaction time. ^ W In step 534, the manager 280 generates a relationship graph. For example, a trend line between the reaction time and the CPU usage rate. In step 538, the manager 28 generates a value Z) of the required reaction time according to the service level agreement, for example, one second, and the value (four) should be Confidence. If the confidence level is 5〇%, you can directly get the CPU usage rate corresponding to this value Z) from equation (7) 26 1287702 玖, invention description. However, for convenience of explanation, this confidence is higher than 50%, such as 95°/. . In step 542, the manager 280 determines a target reaction time value that is faster than one second, such as a value Γ such that when the reaction time is a value of 7, the 95% chance of the reaction 5 time is lower than the value D. In step 546, the manager 280 calculates the spare capacity for the reaction time based on the target value Γ. In step 550, the manager 280 calculates the CPU spare capacity corresponding to the reaction time spare capacity. 10 If step 554 determines that the spare capacity of the CPU is within an acceptable range, indicating that the operation of the system is acceptable, the method then returns to step 502. However, if step 554 indicates that the spare capacity is not within an acceptable range, manager 280 calculates a rate of change in CPU usage in step 558. In step 562, the manager 280 adjusts the operation of the server 110 to adjust the CPU clock speed such that the CPU spare capacity is at an acceptable level. The method then returns to step 502 and continues with the power management process. During the above steps of Figures 5A and 5B, the administrator may be able to close certain applications, merge applications, etc., while meeting certain preset conditions. 20 Power Manager Figure 6 shows an embodiment of a power manager 280 that includes a hypervisor 610, a performance measurement program 620, a performance control program 630, a performance optimization program 640, and program scripts 650 and 660, all communicating via bus 6050. In one embodiment, program 610 is executed in the form of a Windows service or Unix program 1287702, and includes a management user interface 6150 and an initiator 6160. User interface 6150 allows the user to enter assembly information, execute instructions, scripts, and/or make selections in a graphical user interface (GUI). The launcher program 6160 launches the performance measurement program 620, the performance control program 63, and the script "on" and processes the initialization work, such as reading the previous input and storing the composition via interface 615. The performance measurement program 620 monitors system resource performance and parameters included in the service level agreement. The performance program 620 calculates the statistical mean and standard deviation of the performance parameters and compares which ones are necessary based on the service level agreement. Equation 10 620 also calculates the spare capacity value. The program 62 continuously measures and records these parameters via appropriate component and application interfaces, and when necessary, triggers the control program 630 to perform appropriate actions to manage the power supply. The performance control program 630 uses a power saving method to reduce or increase the power supply. Control program 630 includes the % and 15 interfaces required to control system resources. For example, the control program 630 reduces or increases the effective clock speed of the CPU via the TTC circuit, increases or decreases the speed of the disk drive, or turns the device on or off as necessary. In one embodiment, the performance optimization program 640 establishes a dynamic model of the relationship between system performance and "adjustable" parameters. The program 64 adjusts these parameters according to this model to optimize system performance. As discussed above, once the system is optimized, the measurement program 620 and the control program 63 will apply a variety of power savings discussed above at the appropriate time. technology. Scripts 650 and 660 execute power management policies based on primary conditions and requirements. In one embodiment, script 650 is executed during normal operation to implement logic for collecting 28 1287702 defects, invention instructions, and analyzing performance information. For example, the script 65 triggers the statistical average of the performance metrics 620 and the standard deviation, the measured spare capacity, and the like. The script 650 triggers the control program 63 to control the power based on the result of the measurement. 5 Script 660 provides logic to handle exception health. The script 660 closes or starts the corresponding program based on the grouping information. For example, when a "power shortage" condition is encountered, the script 660 triggers the control program 630 to reduce power. The script 660 also closes unimportant programs, etc. Instructions for executing the power manager 10 Figure 7 A flowchart of executing a power management method in accordance with an embodiment is illustrated. In step 704, the system administrator installs a power manager on the server 11A. In step 708, the administrator assembles various information via the interface 6150, 15 for example daily. The time at which the program and service are executed on the server 110. The manager specifies a program that is not necessary but needs to be executed if the situation permits, a program that is not necessary to execute but is executed as required, etc. The manager planning application can be closed, having The lower priority or as needed may be performed, etc. The manager will program the script 650 to normal condition and plan the script 660 to be an exception 20. The manager plans to execute the script when the situation is met. The manager can This information and plan are frequently changed at step 708. In step 712, after the system is powered on, the program 61 performs normal operations. The program 610 also starts the program 620 and the program 630 to measure and optimize the system performance. 29 1287702 Rose, Invention Description In step 716, when the external environmental conditions change, the administrator uses the interface 6150 to send an instruction to notify the power supply. These changes are made by the power manager 28. The power manager 280 executes the scripts 650 and 660 for these new situations via the program 61 in accordance with these new conditions. In the above detailed description, the present invention is made with reference to the specific exemplary embodiments. However, there may be various modifications and changes without departing from the main spirit and scope of the present invention. Therefore, the detailed explanations and illustrations are for illustrative purposes only and are not intended to be limiting. [Simplified Schematic] Figure 1 shows A network that may be implemented by an embodiment of the invention. Figure 2 shows an exemplary computer that may be implemented in accordance with an embodiment of the present invention. Figure 3A shows a look-up table used to find the probability given a parameter. Figure 3B shows The matching rate is used to find the comparison table of the parameters. Figure 4 shows the trend curve of two variables, especially the fifth. Figure 5A shows the management system power according to an embodiment. Figure 5B is a flow diagram illustrating the steps of managing system power associated with SLA parameters in accordance with an embodiment. Figure 6 illustrates power management in accordance with an embodiment. Figure 7 is a diagram illustrating power execution in accordance with an embodiment. Management flow chart 30 1287702 玖, invention description [the main component representative symbol table of the figure] 100···Network 252...World Wide Web 110···Server 256···Server 120-1,102- 2,102-N ...client 150···communication link 200···computer system 204··· central processing unit (CPU) 208...random access memory (RAM) 212···read-only memory (ROM) 216···Storage device 220···Communication interface 224···Bus line 228···Input device 232···Display device 236···Voice control 240...Network 248...Printer 280· · Manager 400... Trend Lines 502, 504, 506, 508, 512, 516, 518, 520, 524, 526, 528, 530, 534, 538, 542, 546, 550, 554, 558, 562 " Step 610 · · Management Program 620 · · Effectiveness Measurement Program 630 · · Performance Control Program 640... Performance Optimization Program 650, 660 · · · Program Script 704,708,712,716 ~Steps 6050... Bus 6150... User Interface 6160... Launcher
3131
Claims (1)
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US10/132,931 US7222245B2 (en) | 2002-04-26 | 2002-04-26 | Managing system power based on utilization statistics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| TW200305800A TW200305800A (en) | 2003-11-01 |
| TWI287702B true TWI287702B (en) | 2007-10-01 |
Family
ID=29248869
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| TW091135992A TWI287702B (en) | 2002-04-26 | 2002-12-12 | Managing system power |
Country Status (2)
| Country | Link |
|---|---|
| US (1) | US7222245B2 (en) |
| TW (1) | TWI287702B (en) |
Families Citing this family (44)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1379035A4 (en) * | 2001-04-09 | 2005-10-19 | Ntt Docomo Inc | Communication data amount measurement system |
| US7017059B2 (en) * | 2003-12-12 | 2006-03-21 | Cray Canada Inc. | Methods and apparatus for replacing cooling systems in operating computers |
| US7406691B2 (en) | 2004-01-13 | 2008-07-29 | International Business Machines Corporation | Minimizing complex decisions to allocate additional resources to a job submitted to a grid environment |
| US7562143B2 (en) * | 2004-01-13 | 2009-07-14 | International Business Machines Corporation | Managing escalating resource needs within a grid environment |
| US7552437B2 (en) * | 2004-01-14 | 2009-06-23 | International Business Machines Corporation | Maintaining application operations within a suboptimal grid environment |
| US20060048157A1 (en) * | 2004-05-18 | 2006-03-02 | International Business Machines Corporation | Dynamic grid job distribution from any resource within a grid environment |
| US7266547B2 (en) | 2004-06-10 | 2007-09-04 | International Business Machines Corporation | Query meaning determination through a grid service |
| US7584274B2 (en) * | 2004-06-15 | 2009-09-01 | International Business Machines Corporation | Coordinating use of independent external resources within requesting grid environments |
| JP4180638B2 (en) * | 2004-10-28 | 2008-11-12 | 富士通株式会社 | Analysis method and apparatus |
| US7308591B2 (en) * | 2004-12-16 | 2007-12-11 | International Business Machines Corporation | Power management of multi-processor servers |
| US7502948B2 (en) | 2004-12-30 | 2009-03-10 | Intel Corporation | Method, system, and apparatus for selecting a maximum operation point based on number of active cores and performance level of each of the active cores |
| US20060149652A1 (en) * | 2005-01-06 | 2006-07-06 | Fellenstein Craig W | Receiving bid requests and pricing bid responses for potential grid job submissions within a grid environment |
| US7793308B2 (en) * | 2005-01-06 | 2010-09-07 | International Business Machines Corporation | Setting operation based resource utilization thresholds for resource use by a process |
| US7707288B2 (en) * | 2005-01-06 | 2010-04-27 | International Business Machines Corporation | Automatically building a locally managed virtual node grouping to handle a grid job requiring a degree of resource parallelism within a grid environment |
| US7761557B2 (en) * | 2005-01-06 | 2010-07-20 | International Business Machines Corporation | Facilitating overall grid environment management by monitoring and distributing grid activity |
| US7668741B2 (en) * | 2005-01-06 | 2010-02-23 | International Business Machines Corporation | Managing compliance with service level agreements in a grid environment |
| US7590623B2 (en) * | 2005-01-06 | 2009-09-15 | International Business Machines Corporation | Automated management of software images for efficient resource node building within a grid environment |
| US7571120B2 (en) | 2005-01-12 | 2009-08-04 | International Business Machines Corporation | Computer implemented method for estimating future grid job costs by classifying grid jobs and storing results of processing grid job microcosms |
| US7562035B2 (en) | 2005-01-12 | 2009-07-14 | International Business Machines Corporation | Automating responses by grid providers to bid requests indicating criteria for a grid job |
| US7319908B2 (en) * | 2005-10-28 | 2008-01-15 | Microsoft Corporation | Multi-modal device power/mode management |
| US7752468B2 (en) * | 2006-06-06 | 2010-07-06 | Intel Corporation | Predict computing platform memory power utilization |
| US8001407B2 (en) * | 2006-10-31 | 2011-08-16 | Hewlett-Packard Development Company, L.P. | Server configured for managing power and performance |
| US8117478B2 (en) | 2006-12-29 | 2012-02-14 | Intel Corporation | Optimizing power usage by processor cores based on architectural events |
| US7386851B1 (en) * | 2008-01-04 | 2008-06-10 | International Business Machines Corporation | System and method for implementing dynamic lifetime reliability extension for microprocessor architectures |
| US20090199019A1 (en) * | 2008-01-31 | 2009-08-06 | Nokia Corporation | Apparatus, method and computer program product for reducing power consumption based on relative importance |
| JP5212381B2 (en) * | 2008-01-31 | 2013-06-19 | 日本電気株式会社 | FEED FORWARD CONTROL METHOD, SERVICE PROVIDED QUALITY CONTROL DEVICE, SYSTEM, PROGRAM, AND ITS RECORDING MEDIUM |
| US8024594B2 (en) * | 2008-03-31 | 2011-09-20 | Intel Corporation | Method and apparatus for reducing power consumption in multi-channel memory controller systems |
| US9253560B2 (en) * | 2008-09-16 | 2016-02-02 | Personics Holdings, Llc | Sound library and method |
| US9026640B2 (en) * | 2009-06-03 | 2015-05-05 | Microsoft Technology Licensing, Llc | Determining server utilization |
| US8904394B2 (en) | 2009-06-04 | 2014-12-02 | International Business Machines Corporation | System and method for controlling heat dissipation through service level agreement analysis by modifying scheduled processing jobs |
| US9836783B2 (en) | 2009-07-24 | 2017-12-05 | Avago Technologies General Ip (Singapore) Pte. Ltd. | Method and system for content selection, delivery and payment |
| US8832483B1 (en) | 2009-12-16 | 2014-09-09 | Applied Micro Cicuits Corporation | System-on-chip with power-save mode processor |
| US9245246B2 (en) * | 2010-04-22 | 2016-01-26 | International Business Machines Corporation | Capacity over-commit management in resource provisioning environments |
| US8452992B2 (en) * | 2010-06-29 | 2013-05-28 | Stmicroelectronics S.R.L. | Power management using constraints in multi-dimensional parameter space |
| US8760713B2 (en) | 2010-06-30 | 2014-06-24 | Konica Minolta Laboratory U.S.A., Inc. | Controlling printer energy consumption |
| US8391174B2 (en) * | 2010-07-13 | 2013-03-05 | Hewlett-Packard Development Company, L.P. | Data packet routing |
| US9256900B2 (en) * | 2010-11-15 | 2016-02-09 | International Business Machines Corporation | Managing service demand load relative to infrastructure capacity in a networked computing environment |
| US20130197895A1 (en) * | 2012-01-31 | 2013-08-01 | Zhikui Wang | Real-time server management |
| KR101976397B1 (en) | 2012-11-27 | 2019-05-09 | 에이치피프린팅코리아 유한회사 | Method and Apparatus for service level agreement management |
| US9575542B2 (en) * | 2013-01-31 | 2017-02-21 | Hewlett Packard Enterprise Development Lp | Computer power management |
| CN105468823B (en) | 2015-11-16 | 2019-01-25 | 广州广电运通金融电子股份有限公司 | A kind of self-service device energy-saving control method and device |
| US20210157384A1 (en) * | 2018-07-20 | 2021-05-27 | Hewlett-Packard Development Company, L.P. | Assigning power sources |
| CN111382163B (en) * | 2018-12-27 | 2023-03-21 | 技嘉科技股份有限公司 | Efficiency management system, method for providing and updating efficiency parameter and storage medium |
| KR20200101210A (en) * | 2019-02-19 | 2020-08-27 | 삼성전자주식회사 | Electronic device and method for determining operating frequency of processor |
Family Cites Families (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA1329432C (en) * | 1988-11-02 | 1994-05-10 | William Davy | Method of memory and cpu time allocation for a multi-user computer system |
| FR2727540B1 (en) * | 1994-11-30 | 1997-01-03 | Bull Sa | LOAD DISTRIBUTION ASSISTANCE TOOL FOR A DISTRIBUTED APPLICATION |
| US5699412A (en) * | 1995-09-18 | 1997-12-16 | Intervoice Limited Partnership | Systems and methods for statistical distribution of messages in a message recording system |
| US5872972A (en) * | 1996-07-05 | 1999-02-16 | Ncr Corporation | Method for load balancing a per processor affinity scheduler wherein processes are strictly affinitized to processors and the migration of a process from an affinitized processor to another available processor is limited |
| US6078943A (en) * | 1997-02-07 | 2000-06-20 | International Business Machines Corporation | Method and apparatus for dynamic interval-based load balancing |
| JPH1118120A (en) * | 1997-06-20 | 1999-01-22 | Fujitsu Ltd | Process control system for intelligent network system |
| US5958058A (en) * | 1997-07-18 | 1999-09-28 | Micron Electronics, Inc. | User-selectable power management interface with application threshold warnings |
| US6128642A (en) * | 1997-07-22 | 2000-10-03 | At&T Corporation | Load balancing based on queue length, in a network of processor stations |
| US6317786B1 (en) * | 1998-05-29 | 2001-11-13 | Webspective Software, Inc. | Web service |
| US6687735B1 (en) * | 2000-05-30 | 2004-02-03 | Tranceive Technologies, Inc. | Method and apparatus for balancing distributed applications |
| US6597907B1 (en) * | 2000-05-05 | 2003-07-22 | Ericsson Inc. | Detection of a deadlocked resource condition in a pool of shared resources |
| US6738933B2 (en) * | 2001-05-09 | 2004-05-18 | Mercury Interactive Corporation | Root cause analysis of server system performance degradations |
| US20030055969A1 (en) * | 2001-09-17 | 2003-03-20 | International Business Machines Corporation | System and method for performing power management on a distributed system |
| US6898718B2 (en) * | 2001-09-28 | 2005-05-24 | Intel Corporation | Method and apparatus to monitor performance of a process |
-
2002
- 2002-04-26 US US10/132,931 patent/US7222245B2/en not_active Expired - Fee Related
- 2002-12-12 TW TW091135992A patent/TWI287702B/en active
Also Published As
| Publication number | Publication date |
|---|---|
| US20030204758A1 (en) | 2003-10-30 |
| TW200305800A (en) | 2003-11-01 |
| US7222245B2 (en) | 2007-05-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| TWI287702B (en) | Managing system power | |
| TWI274986B (en) | Method and system for managing power of a system, and computer-readable medium embodying instructions for managing power of a system | |
| Meisner et al. | Power management of online data-intensive services | |
| Femal et al. | Boosting data center performance through non-uniform power allocation | |
| CA2522467C (en) | Automated power control policies based on application-specific redundancy characteristics | |
| US7308591B2 (en) | Power management of multi-processor servers | |
| US10289184B2 (en) | Methods of achieving cognizant power management | |
| US8104041B2 (en) | Computer workload redistribution based on prediction from analysis of local resource utilization chronology data | |
| US8527997B2 (en) | Energy-aware job scheduling for cluster environments | |
| US20050177755A1 (en) | Multi-server and multi-CPU power management system and method | |
| GB2437846A (en) | Power Allocation Management in an Information Handling System | |
| Femal et al. | Safe overprovisioning: Using power limits to increase aggregate throughput | |
| US20090327775A1 (en) | User imposed power constraints on web server based on user preferences | |
| Yang et al. | Dynamic cluster reconfiguration for energy conservation in computation intensive service | |
| Wolski et al. | QPRED: Using quantile predictions to improve power usage for private clouds | |
| Imada et al. | Power management of distributed web savers by controlling server power state and traffic prediction for QoS | |
| US20230229216A1 (en) | System and methods for server power management | |
| Milenkovic et al. | Power-aware management in cloud data centers | |
| Menascé | Modeling the tradeoffs between system performance and CPU power consumption | |
| Liu et al. | Queuing theoretic analysis of power-performance tradeoff in power-efficient computing | |
| Sen et al. | Frequency governors for cloud database OLTP workloads | |
| Tian et al. | Energy Management | |
| Lynar et al. | Reducing grid energy consumption through choice of resource allocation method | |
| Leh et al. | Model-Driven Joint Optimization of Power and Latency Guarantee in Data center Applications | |
| AlLee | Green Microprocessor and Server Design |